What to measure in digital marketing to make decisions

More visits, clicks or contacts do not always mean better results. Learn which indicators your company needs, how to interpret them and how to turn data into improvement decisions, with simple formulas and a practical example.

Your monthly marketing report arrives: website visits have grown by 30% and campaigns generate more clicks. Yet the sales team receives enquiries less suited to the offering and takes longer to respond.

Is the business improving, or has digital activity merely increased?

Figures can be correctly calculated and still fail to answer that question. A visit, a request and a purchase represent different moments. To interpret them, we need to know what the company wants to achieve, what happens between those moments and whether it can handle the opportunities received.

Measurement begins with a decision: what we need to understand to maintain, correct, investigate or develop an initiative. Then we choose indicators and check their limitations.

Measuring is not about celebrating growing figures. It means using sufficient, reliable information to direct digital assets better.

1. What measuring digital marketing for decisions means

Measuring digital marketing means observing information about our actions, assets and people’s responses to understand their contribution to business objectives.

We can measure page arrivals, enquiries received, confirmed bookings or an initiative’s development cost. But these figures do not all mean the same thing.

A visit indicates a recorded website interaction. A request may show someone decided to contact us. A purchase confirms a commercial outcome. A properly fulfilled booking also represents an operational responsibility. They are related but different events.

The mistake of starting with tools

A company may use Google Analytics, Search Console, advertising tools, social media statistics and commercial management software. Each system offers its own reports.

The problem arises when we build a dashboard simply by gathering the figures these tools present. We end up with visits, sessions, clicks, impressions, percentages, costs and events, but perhaps nobody has defined the questions to answer.

Imagine a company wants more direct bookings. Knowing whether website visits are growing is insufficient. It must also understand whether visits reach appropriate pages, people find availability and ultimately manage to book.

Now imagine another company receives enough bookings but struggles to manage them. Its priority may differ: errors, pending enquiries, cancellations or incident resolution time may need observing. The same website may need different indicators depending on the problem to solve.

First the objective, then the indicators

A simple starting point is three questions:

What do we want to achieve?
The outcome or capability that matters to the business.

What do we need to observe?
Signals that help us understand whether we are progressing.

What decision could we make?
The intervention, investigation or continuation depending on that information.

For example:

This approach has an important advantage: it prevents adopting indicators simply because another company uses them.

A restaurant’s needs differ from those of a professional services company. So do those of a new business and an established one.

The British government’s digital service measurement guide proposes starting with service purpose, establishing objectives, forming hypotheses and then selecting the data needed to test them. [1] Marketing planning connects those questions with business initiatives.

No universal indicator list can direct every business properly. We must build a relationship between objectives, information and decisions.

2. Which digital marketing metrics a company actually needs

The answer depends on what we want to achieve. But we can organise common indicators by the business needs they help investigate.

The aim is not to measure everything in the following table. Its usefulness lies in recognising what may be relevant to our situation.

Indicators by business objective

Some indicators describe activity; others observe people’s responses or approach business outcomes. We need to distinguish them.

Visibility does not necessarily mean opportunity

A company can appear thousands of times in Google results. That represents visibility, but does not demonstrate those appearances relate to the services it wants to sell or that people visited its website.

Impressions and clicks help study search presence. We then need to observe what happens to people who arrive.

A request is not a customer

Imagine a website receives twenty forms. Has it obtained twenty commercial opportunities? It depends.

Some may be duplicates. Others may be job applications, supplier enquiries or requests for services the company does not offer.

We need to define a valid commercial request and, if the business model requires it, when a request is qualified.

For example, a private events company may consider it relevant that the contact requests a service available in its territory, for a date and guest count compatible with its offering.

Criteria must be established before results are evaluated. Otherwise, we may change them whenever a figure looks unfavourable.

A purchase does not automatically mean profitability

A confirmed sale or booking is a commercial result. But we still need its economic value and the resources needed to obtain and fulfil it.

For businesses with different services, two purchases may represent very different revenues and margins. Results must therefore be interpreted according to the business model.

A campaign generating many requests is not necessarily the most attractive. Another may produce fewer contacts with a better fit and proportionate cost.

Measurement should help recognise these differences, rather than reduce the whole business to a traffic or contact figure.

3. Three levels connecting marketing and results

At PhiStudio, we can organise measurement into three connected levels:

Outcome → Progression → Health and capacity.

They are not three departments or separate reports. They are perspectives to connect to understand events.

Level 1. Outcome: what we are achieving

The first level observes what matters to the business: a confirmed booking, enrolment, accepted commercial opportunity or fewer incidents. It can also be an operational improvement, such as customers completing a task without assistance.

Not every outcome must immediately be expressed in euros, but it must relate to a recognisable business function.

Imagine a company has developed a new service page. Its objective may be better-informed commercial requests. Increased visits may be interesting, but the outcome to evaluate concerns requests and their quality.

Level 2. Progression: where people advance or stop

The second level studies what happens before the outcome.

Someone may discover an offering, visit a page, check terms, request information and ultimately purchase. But not every customer follows the same journey. Some arrive after a referral, others research for weeks, and others phone directly.

We need not impose an identical funnel on everyone. We must identify the steps that actually matter to each business.

For example:

These signals can locate difficulties.

If visits grow but requests do not, demand fit or page experience may need investigating. If requests are suitable but rarely become purchases, the problem may be further along: offering, price, availability or commercial follow-up.

Data will not necessarily provide an automatic explanation, but can help formulate better questions.

Level 3. Health and capacity: what enables the system to function

This level often falls outside traditional marketing reports, yet is fundamental.

A company may attract suitable opportunities and lose them because it lacks response capacity. It may also record forms incorrectly, have broken links or use data whose quality has not been checked.

We therefore need to observe operating conditions. For example:

• Request response time.
• Form or booking system errors.
• Availability of essential functions.
• Pending requests.
• Team capacity to meet demand.
• Quality and consistency of records used.

These signals help interpret results.

Imagine a campaign produces more requests than the team can handle. Increasing investment further could worsen matters. Operational capacity may need improvement first.

Marketing can generate opportunities. The business must be ready to handle them.

How the three levels relate

Outcome shows what the business achieves. Progression locates where people advance or stop. Health and capacity indicates whether the website, data and team can sustain the journey.

Together, they distinguish real improvement from increased activity that has yet to become value.

4. Defining a useful indicator: objective, formula and context

Before interpreting data, we must define what we measure.

This may seem minor, but two people can use the same word for different things. One may call any form submission a “conversion”, another only valid commercial requests, and a third confirmed sales.

If the label is identical but events counted differ, reports become difficult to compare.

Four concepts worth distinguishing

Metric: a specific measure, such as visits, requests or bookings.

Indicator: a measure selected to observe a relevant function or objective.

Baseline: the starting point used as a reference.

Target: the outcome sought within defined conditions and a period.

For example:

We need not turn every available figure into a priority indicator. Nor should we invent a target merely to display a goal on the dashboard.

A useful target considers the starting point, available resources, historical behaviour and uncertainty.

Five calculations worth understanding

Formulas are simple; what matters is defining what we count and comparing equivalent periods.

A session is not necessarily a person. A valid request does not guarantee a sale. An open opportunity should not be counted as lost before a reasonable decision period ends.

For cost indicators, clarify whether only advertising or also management, production and tools are included. Advertising cost per request differs from the full cost of acquiring a customer.

The following example shows how to use these relationships without confusing them with definitive business outcomes.

The correct formula needs a correct definition

Mathematical operations can be simple. Difficulty arises when deciding what we count.

Each important indicator should therefore retain a clear definition of its unit, source, period and limitations.

A poorly defined formula can produce perfectly calculated figures that lead to wrong decisions.

A digital conversion is not necessarily a sale

This distinction is fundamental.

Google Analytics 4 allows certain important business actions to be marked as key events, such as a form submission or another relevant site action.

Google distinguishes these from conversions used to measure and optimise advertising performance in Google Ads, although the concepts can be technically connected. [2]

But a tool recording an important action does not mean the business has acquired a customer.

We should differentiate:

Not every business needs all these stages recorded, but it must know which it observes.

A tool records events. We must define their business meaning.

5. Practical example: more visits, fewer enrolments

Let us connect the indicators through a complete case.

Imagine a small in-person training centre offering specialist courses for adults. Its website presents programmes and allows information requests.

In recent months it has worked to improve visibility and attract more interested people. Management receives a positive report: course-page visits have risen substantially.

Yet the sales team feels enquiries are less suitable and harder to turn into enrolments. We need to investigate.

This example is entirely hypothetical. Figures, periods and situations are educational inventions. They do not represent actual PhiStudio results or permit attributing changes to a specific initiative.

Step 1. We define what we want to achieve

The business objective is not simply more visits. The centre wants more enrolments from people suited to its programmes.

To understand events, we must observe visitor arrivals, requests received, commercial fit, confirmed enrolments and the team’s capacity to handle them.

Step 2. We establish comparable periods

Imagine studying two eight-week acquisition periods. To simplify, assume reasonably comparable seasonal and commercial conditions and unchanged measurement definitions.

We also establish an important condition: enrolments are counted within 30 days after each request, comparing only groups for which that period has elapsed.

This avoids treating a recent request as a failure before it has time to become an enrolment. It does not mean every enrolment decision occurs within thirty days; we simply use the same comparison window.

Step 3. We gather the data

The data seem to tell different stories. Let us interpret them.

First reading: visibility increases

Sessions move from 1,000 to 1,500: 50% growth.

Recorded session volume has clearly improved. But we do not yet know whether new visits relate to courses the centre needs to sell.

They could come from informative searches, campaigns targeting different audiences or sources previously less prominent.

The next question should be:

Where do these visits come from and which needs do they represent?

Second reading: requests increase much less

Valid requests rise from 40 to 45, an increase of 12.5%.

But the request rate against studied sessions changes.

Period A:

40 ÷ 1,000 × 100 = 4%.

Period B:

45 ÷ 1,500 × 100 = 3%.

The rate falls by one percentage point. We have more visits and five extra requests, but fewer sessions proportionately generate requests.

Does that mean the website works worse? Not necessarily.

Visit composition may have changed. More people may now seek general information and not yet be ready to request a course. There could also be a messaging or experience problem.

We must investigate before concluding.

Third reading: commercial fit declines

Here is a more concerning signal for the defined objective.

In period A, 24 of 40 requests qualify: 60%. In period B, 18 of 45 meet the criteria: 40%.

The proportion falls twenty percentage points. The absolute qualified request count also moves from 24 to 18: a 25% reduction.

We now have a fuller picture: visits rose substantially, requests rose slightly, but suitable requests fell.

This may indicate part of new acquisition is insufficiently aligned with the offering.

However, we must also check consistent qualification criteria. The evaluator may have changed or admission requirements become stricter. Without equivalent definitions, comparison weakens.

Fourth reading: enrolments fall

The centre confirms 10 enrolments from the first request group during their respective thirty-day follow-up windows, and 7 from the second.

That is a 30% decrease: a less favourable commercial outcome, but not yet an explanation.

It may relate to fewer qualified requests. Prices, places available, course schedules or purchase terms may also have changed. There may even be differences within the thirty-day window.

We should not automatically attribute the decline to digital acquisition.

Fifth reading: the team takes longer to respond

Median first response time rises from four to twenty hours.

This means the request in the middle of the ordered response times received a later reply in the second period.

The median is useful because extreme times affect it less than a simple average. But it does not explain the whole distribution: check whether some requests receive immediate responses while others wait days.

The change deserves attention. The team may have more work, the procedure may have changed or coordination problems may exist.

What matters is recognising that response capacity forms part of the commercial journey, even though it is not an advertising metric.

Step 4. We interpret the three levels together

We can summarise the situation.

We now understand why the initial report was insufficient. It showed improved visibility while business outcomes and other relevant signals moved differently.

Step 5. We identify what we still do not know

Before recommending changes, investigate several possibilities.

The origin of new sessions.

Have channels changed? Do they come from new campaigns or informative queries?

Request content.

Do people ask about courses we actually offer? Are there recurring questions?

The commercial offering.

Have prices, timetables, availability or requirements changed?

Follow-up.

Does every request receive a response? Are there relevant differences between responsible people or channels?

The time comparison.

Are the periods sufficiently comparable? Are there seasonal factors or significant external changes?

These questions define a diagnosis. We have not yet demonstrated what causes the enrolment decline.

Step 6. We make a proportionate decision

Given available information, I would not recommend immediately increasing acquisition investment, nor rebuilding the website.

The initial decision could be:

Investigate the origins and fit of new requests, review the response process and postpone investment expansion until results are better understood.

This can become concrete actions: review queries and channels generating visits; check qualification criteria; analyse request management and response delays; and correct confirmed problems before developing new initiatives.

Then we would observe indicators again, not to prove our hypothesis correct, but to check what changed and what we can learn.

What the example teaches

A company can increase visibility without improving commercial results. It may even get more contacts yet fewer suitable opportunities.

But neither should every decline be treated as proof of failure. We need definitions, periods, data quality and contributing factors understood.

Figures do not decide for us. They help us ask questions, recognise problems and choose better-founded interventions.

6. How to avoid wrong conclusions

Proper measurement is more than selecting indicators and calculating percentages. We must know how much to trust them.

A figure may be correctly calculated yet insufficient to justify a decision. Context, data quality and uncertainty shape every interpretation.

Compare periods that make sense

Imagine a tourism company in the Val d’Aran. It receives many requests in August; demand falls in November.

Does that mean marketing worsened? Not necessarily. Seasonality can explain much of the difference. Comparing August with November without considering the tourism calendar can mislead.

Some businesses should study equivalent periods across seasons. We must also consider changes in prices, offering, budget and channels. Even the same months in consecutive years require caution if business conditions changed.

We are not seeking identical periods, which rarely exist. We seek sufficiently comparable references and recognise their limits.

Observe absolute quantities alongside percentages

Percentages aid comparison, but can exaggerate growth impressions with few records.

Imagine four commercial requests one month and six the next. The change is 50%, but only two extra requests.

There may be real improvement or ordinary fluctuation. Small samples require restrained conclusions. Sometimes a longer period helps; elsewhere, retain uncertainty and wait before major changes.

GOV.UK guidance recommends context, trends over time, data quality checks and appropriate statistical methods where relevant. [1]

Not every commercial decision needs a statistical study. But the more limited the evidence, the more cautious our conclusions should be.

Check that tools count the same things

Imagine Google Analytics records thirty-five form submissions, but the commercial record contains twenty-nine requests.

Is there an error? Possibly. First determine what each figure represents.

There may be repeat submissions, tests, date differences, filters or validation criteria. Sometimes systems record different events.

Google explains that its own Analytics reports can differ due to filters, thresholds, modelling or processing times. [3]

Search Console and Analytics also use different methods and units. Search Console measures, among other things, clicks from Google results; Analytics may measure website sessions.

A click and session are not exactly the same event, so counts need not match. [4]

We should not force artificial agreement between tools. We should understand each datum.

Five checks before interpreting a change

When an indicator changes significantly, review:

A simple review can prevent hasty decisions.

Combine quantitative and qualitative information

Not everything we need appears in a chart.

Customer conversations may reveal questions analytics misses. Sales staff may notice wrong expectations in requests. Service staff may identify recurring process errors.

These observations complement rather than replace quantitative data. They help form hypotheses and determine what to investigate.

If digital indicators improve while the sales team reports declining request quality, neither perspective should be dismissed automatically.

Contradiction is a reason to investigate, not select the data that suit us.

Correlation does not mean causation

This principle is fundamental.

Imagine a website renewal followed by more requests in subsequent weeks. Can we claim new design caused growth? Not yet.

It may coincide with high season, another company’s recommendation or increased advertising investment.

Redesign may have contributed, but consecutive events do not prove causation.

Where volume and conditions allow, more controlled tests can be designed. Elsewhere, we work with imperfect observations and comparisons.

Clearly distinguish:

Observation: what happened.

Hypothesis: what might explain it.

Pending evidence: what needs checking.

For example: requests rose after a page change. We believe new presentation may have helped people understand the offering. But other changing factors need investigation.

This does not prevent decisions. It enables decisions recognising uncertainty.

Responsible measurement also respects privacy

A company need not track every visitor movement to understand whether its assets work. Measurement should be proportionate to the questions answered.

Technologies must also meet corresponding obligations.

In Spain, the Agencia Española de Protección de Datos’s cookie guide establishes information and consent requirements and certain exemptions depending on purpose and usage conditions. [5]

We cannot assume a tool is automatically exempt because it serves analytics, nor that every measurement method has identical conditions. The specific technology and processing need assessment.

Where digital information has partial coverage for technical or privacy reasons, legitimate commercial records, aggregate data and operational observations can complement it.

That limitation is not an invitation to collect unnecessary personal information.

Good measurement captures sufficient, reliable and properly obtained evidence, rather than the most data.

7. Connecting indicators with economic results

A company needs to understand what its investments produce. But another common confusion must be avoided.

Economic indicators differ. Cost per contact is not cost per customer. Revenue is not necessarily profit. Platform-calculated advertising return does not automatically demonstrate how much additional money the business earned because of a campaign.

Cost per request does not tell the whole story

Imagine two campaigns. The first produces twenty requests at €10 advertising cost per request. The second generates ten at €18.

The first appears more efficient, but request quality is still unknown. It may mainly attract people wanting services the company does not offer, while the second produces fewer better-fitting contacts.

We must also consider what follows: how many requests are handled, how many become purchases, their revenue and service delivery costs.

Without that information, cost per request gives an incomplete view.

What ROAS means

ROAS, or return on advertising spend, expresses the relationship between conversion value attributed to advertising and its cost.

The usual calculation is:

ROAS = Attributed conversion value ÷ Advertising spend

Imagine €500 campaign spend and €2,000 conversion value attributed by the platform.

2,000 ÷ 500 = 4.

This can be expressed as 4:1 ROAS or 400%.

But it does not mean €1,500 profit. We must distinguish attributed value from confirmed revenue and consider relevant costs.

Google Ads defines conversion value/cost as an advertising performance metric, while profitability assessment requires additional financial information. [6]

Revenue, costs and economic contribution

Suppose that €2,000 represents real sales. Depending on business and analysis, contribution requires considering product or service costs, advertising investment, relevant management and production costs, tools and related resources.

Different companies have different economic structures. We need not impose one profitability formula on every project, but must avoid presenting revenue as profit.

Attribution alone does not prove causality

Imagine someone sees an ad, then researches on Google, receives a friend’s recommendation and purchases.

Which action caused the sale? The answer is not necessarily evident.

An advertising platform may attribute value according to its attribution model and rules. That does not automatically prove the whole sale would have been lost without the ad.

Studying incremental campaign effects requires methods estimating additional results caused by advertising. Rigorous analysis is not always possible, especially for small companies with few purchases.

In those situations, avoid strong claims about attributed profitability. ROAS and other indicators can guide decisions, provided their limits are understood.

The aim is assessing sustainability

From External Digital Direction, what matters is not making an advertising figure attractive. It is understanding whether the initiative sustainably contributes to the business.

A campaign can generate sales yet remain unattractive because of costs. It may also support long-term objectives not immediately reflected in purchases.

Measurement must therefore connect with strategy and the economic model.

The aim is not proving marketing works because it produces activity. It is understanding its contribution and whether resources are justified.

8. Turning data into improvement decisions

We have identified indicators, calculated and interpreted results. The most important step remains:

What do we do with what we have learned?

A professional review should do more than produce a report. It should identify what needs attention, what to maintain and which information remains missing.

The measurement and improvement cycle

A simple sequence is:

Question → Indicator → Context → Interpretation → Decision → Action → New review.

Each stage has a purpose. The question establishes what to understand. The indicator provides information. Context enables proper interpretation. Interpretation distinguishes observations, hypotheses and limitations. The decision determines what to do. Action turns it into concrete work. The next review checks subsequent events.

This cycle does not guarantee continuous improvement, but provides a shared reference and supports learning from interventions.

Which decisions we can make

A review can reach different conclusions.

Maintain. When the system fulfils its purpose and there is insufficient reason to modify it.

Correct. When an identified local problem deserves attention.

Investigate. When data contradict or causes remain unclear.

Test. When a specific hypothesis can be examined through proportionate intervention.

Retire. When an initiative loses its function or no longer justifies its resources.

Wait and observe. When available information cannot support a responsible recommendation.

None is necessarily superior. Choose what best fits evidence and business priorities.

An applied decision record

Return to our training centre. After reviewing indicators, we could document:

This record demonstrates something important. We have not automatically turned data into a new campaign or promised a particular improvement.

We identified a problem, formed hypotheses and established a proportionate decision. We now know what to investigate, who is responsible and when results will be reviewed again.

Not every meeting should end with more tasks

Imagine reviewing a website. Available indicators show stable operation, requests fit well, the team responds properly and no new needs emerged.

Preserving what works may be appropriate. We need not change design, alter the form or launch another campaign merely to demonstrate activity.

Data may also remain insufficient. We can wait for a comparable period, provided waiting has a reason and review point.

Continuous improvement does not mean continuous intervention. It means retaining the ability to observe, understand and decide when intervention is worthwhile.

A minimal template to begin

You do not need a complex dashboard. Start with a tracking sheet gathering essential decisions.

The system can expand as needed, but should not become a disproportionate administrative burden.

The template’s value is not filling every field. It is giving important decisions an understandable justification that can be reviewed later.

9. Measuring less can help you direct better

A company can accumulate charts and still not know what to do. A useful system connects each objective with a few reliable signals, context and a possible decision. Sometimes the decisive signal lies outside a platform: team response time or a repeated customer question.

When reviewing results, we should answer: what happened, what does it mean for the business, what needs checking and what will we do now? The answer may be maintain, correct, test, retire or wait for better evidence.

Measuring digital marketing means understanding what adds value and deciding the next step, rather than demonstrating activity.